Quantum reinforcement learning during human decision-making.

Area of Science:

• Cognitive Neuroscience
• Quantum Cognition
• Computational Psychiatry

Background:

• Classical reinforcement learning (CRL) is widely used in neuroscience and psychology.
• Quantum reinforcement learning (QRL) shows promise in simulations but lacks empirical testing in humans.
• Existing quantum cognition models lack neuroscientific grounding.

Purpose of the Study:

• To investigate if QRL can adequately explain value-based decision-making in humans.
• To compare the performance of QRL models against CRL models using empirical data.
• To explore the neurobiological basis of QRL in human decision-making.

Main Methods:

• Behavioral and functional magnetic resonance imaging (fMRI) data were collected from healthy and cigarette-smoking participants.
• Participants performed the Iowa Gambling Task.
• Two QRL models and twelve CRL models were compared against the collected data.

Main Results:

• QRL models demonstrated strong performance, comparable to the best CRL models across all participant groups.
• QRL models identified representations of quantum-like internal-state-related variables in the medial frontal gyrus.
• These findings were consistent in both healthy individuals and cigarette smokers.

Conclusions:

• Value-based decision-making in humans can be effectively modeled using QRL at both behavioral and neural levels.
• The study provides empirical evidence linking QRL to human cognition and its neural underpinnings.
• This research bridges the gap between quantum cognition theories and neuroscience.